Go back

How AI is breaking the internet ft. Cory Doctorow

from Blood in the Machine

1h 6m 37s

How AI is breaking the internet ft. Cory Doctorow

The recent wave of global tech outages—spanning Microsoft, the FAA, major banks, and airlines—exposes a deep systemic crisis in digital infrastructure. These failures are not due to AI malfunction but are rooted in decades of corporate austerity, where skilled workers are fired and replaced with AI, leading to brittle, outdated systems. Early digital systems were built on complex, custom code with layers of technical debt, and when merged through mergers and acquisitions, they become vulnerable to collapse. The erosion of human expertise—what researchers call "process knowledge"—means systems lack resilience and fail when stressed. A major factor is the ideology of efficiency over sustainability, where companies prioritize cost-cutting over maintaining stable operations. This has been exacerbated by the rapid push for AI adoption, especially in sectors like finance and healthcare, where AI is promoted as a miracle solution despite evidence of poor performance and high error rates. Critics argue that AI is not solving problems but replacing human judgment with flawed, statistically plausible hallucinations. The current situation reflects a dangerous bubble built on ideological fervor, misaligned incentives, and unchecked automation. If this bubble bursts—triggered by failed IPOs, regulatory resistance, or financial instability—it could lead to widespread job loss, the loss of critical expertise, and a long-term decline in essential services. The crisis is not just technological but social, undermining trust in both institutions and the promises of AI. As seen in the dismissal of AI-related failures and the rebranding of "AI" as "super intelligence" (SI), the narrative is being manipulated to avoid accountability. Ultimately, the collapse of these systems is a symptom of deeper failures in management, labor practices, and the belief that technology alone can solve complex human problems. The coming years may reveal whether this technological overreach leads to a sustainable future or a period of widespread instability and loss.

Transcription

11733 Words, 63809 Characters

English
If you think about like an avalanche, right? So an avalanche is a set of very small, almost undetectable destabilizations where things tremor and shift and become a little more unstable and eventually you get to the point where one more instability added to the system doesn't cause a minute almost undetectable shift. It causes a total catastrophic collapse, right? And so anything that can't go on forever eventually stops. You can't just keep dropping snowflakes onto the unstable mountainside, eventually the combination of heat, rain, temperature fluctuations, more snow is going to make it all fall, right? And so I think this is this is true of the finances, the AI sector, which are brittle from several different directions, right? We learned a new source of prittleness, which is that the people running these firms are insatiably horny and vulnerable to Chinese spies. But like we had all these other ones, right? Like, you know, there's a baroque data center financing deals that are being financed by Emirates, who suddenly need to rebuild their liquid natural gas terminals because Trump decided to get a rant about them into smithereens, right? And they're all like stacked. So if one party pulls out of one deal, 50 deals all collapse. So that could be the thing that does. The internet, as we know it, seems to be falling apart. On Thursday, November 12th, Microsoft suffered a massive global outage, shutting down access to Outlook teams and websites running on Azure for not one, but two days in a row. The very next day, the Federal Aviation Administration system crashed, forcing the agency to ground flights nationwide for just the second time since 9/11. And the hits keep coming. American banks like JP Morgan Chase saw their websites crash, taking payment processors like Zell down with them. United airline system went down, then Apple pay, then who knows what's next on and on. It goes. And what did everyone immediately blame? AI, of course. Why wouldn't we? We've all seen headline after headline blaming rogue AI, rogue AI for corporate hacks and disruptions and many Americans are now accustomed to experiencing AI largely as an irritant and annoyance, a layer of slop coating the internet, a bot they've had to incorporate into their workflow, whether they want it or not. Or a feature that keeps returning dubious results, where reliable web search used to be. Now, it's looking unlikely that AI had much to do with most of these outages, at least not directly. But AI still very much played a role. Take Microsoft. The tech giant has laid off tens of thousands of employees in recent months, citing increased efficiencies from AI, right? Those employees might have well been the ones to prevent or correct the outages that everyone just experienced. And don't forget, the FAA was an early target of the Trump administration and its AI obsessed Elon Musk led Doge effort to gut government agencies and replace them with automation. 400 employees were laid off as part of that effort. But this is not the AI story that Silicon Valley wants anyone to hear. That AI's impact on productivity has been a deeply mixed bag so far that firms are reducing token spend and hiring back employees they laid off for AI. That a majority of Americans are not excited about AI, but concerned. Silicon Valley especially does not want anyone to hear this as anthropic and open AI. Make the final push for their much anticipated IPOs, pitching themselves to investors to you and me to stock buyers everywhere. They're going red in the face, repeating the same utopian pitches for AI that they have for years now. It will cure cancer. It will cure Alzheimer's. It will cure every disease in 10 years. No, if today we look at things like cancer, Alzheimer's, heart disease, mental illness, these will be relics of the past in I think less than a decade. I think if we do this right, we can create an age of miracles where humanity is able to solve problems that have eluded us for millennia and age of miracles. He calls it just over the summer. Sam Altman said AI is about to deliver us a genie that can grant any wish. And we are close to creating a genie that can grant any wish. This is their actual pitch, not exaggerated at all for effect. This is what they are saying. And more than that, this is what they are saying as the FAA system is down as Microsoft servers are down as these negative feelings swirl around the tech world in general. The Trump administration for its part, rather than addressing AI's role in the mass plane grounding, directed reporters to refer to it as super intelligence or SI. There are thoughts that this could be an AI-related error or issue. We call it SI, I remember SI. The use of the word artificial makes intelligence fake. It makes it sound fake and it is not fake. It's actually amazing. So it's super intelligence. In other words, welcome to the new world of super intelligence dash SI. Let's see if that goes. It sounds much better. It is much better and it's much more accurate. It's a branding exercise, right? It's an idea. But the branding exercise is wearing thin. People can see for themselves that systems are not functioning as promised that most of their lives have not improved one iota as a result of AI. And oftentimes have even gotten worse. So what gives? What's really happening with all of our crucial digital infrastructures? Did AI destroy the internet as we know it one way or another? How did we get here and where do we go next? To answer those questions, I am joined by perhaps the man best suited to guide us through our crumbling digital landscape, Cory Docturo, science fiction writer, online activist, and author of books like InShitification. And most recently, a reverse-centre's guide to life after AI. Let's get to it. Well, it's been what you might call a banner week or two for InShitification, right? First, Microsoft suffers a global multi-day outage. Outlook is down. Teams is down. Azure sites are down. Then just to get in on the fun, the FAA entire network goes down. Plains are grounded. Websites not working. Is it AI? Is it a router? Who knows? And then we just let the hits keep coming right into this week. The banking system. A bunch of chase banks, websites go down and then for good measure back to what was it? United Airlines. All crashing. All this global sort of internet outage. And I could not help but think of of your work, Cory, InShitification swirling all around as the world we live in. So I thought, you know, why not call up the spirit guide to the world of InShitification? It's, unfortunately, I think it's an honorific that you're going to have to live with for the last year. Who promote you on my headstone? So in short, you know, he's all this stuff. I mean, this stuff has happened before, but the fact that it's all happening now and happening to such an extent fuels of the moment. So it would, but is it, what's going on? Can you explain a little bit? Is this coincidence or is this indeed sort of a, you know, a nuclear grade example of InShitification indeed? Yeah, I mean, there's definitely some aspect of coincidence here, right? I don't think there was a correlation between air traffic control and, you know, the computers at chase or whatever. But I do think that what all these things have in common, well, there are a few things they have in common. So one is that like from a computer perspective, they're all in the category of digital systems that are especially vulnerable because they started early, right? So like basically things that you computerize early are more likely to have really old, gnarly stuff in them because they started early. So insurance companies banks, the aviation system and so on. They can't be turned off. And so when you upgrade them, you tend to just layer something new on top of them or like maybe you have like an emulator that has the old thing on them. And so there's a lot of craft. And then this isn't really the aviation system, but the other systems definitely are involved a lot of mergers. So you remember the giant credit bureau breach. So it turned out that the credit bureau breach was in Maine, the problem of that credit bureau having bought a lot of other credit bureaus, and aquifax having bought a lot of other credit bureaus. And then like the reason that one firm buys another is to reduce its operating expenses. So what do they do? Why are all the redundant people? So now you have two computer systems that are like old, custom built, crafty, have lots of layers of abstractions sitting on top of them, each of which represents a seam that can rupture. And then you smush them together while firing half the people who understand how they work. And this leaves the system in a really brittle state. And then, you know, the other two things that these systems have in common is that both air traffic control and all these big private sector firms have been ruled by an ideology of austerity and this idea that you, if you have someone who's like not working right at this minute, you should fire them. Rather than keeping them in reserve in case something breaks because, you know, efficiency is the opposite of resilience. If resilience is like, oh, we have a backup for that. Efficiency is, we don't spend money on things we're not using. So, you know, air traffic control, particularly like God, governments hate air traffic controllers. They're skilled, they're powerful. If they go on strike, the whole country stops working. It's really hard to replace them, you know. - Famously. - Yeah, since Lincoln, you know, we've had war on them. So, we have these systems that are sort of brittle because they're old and they, the way that they get assembled. And then this is exacerbated by successive managers, whether those are governments or firms that cut away at the people who know how they work. And then you have like this final coup de grove, which is like firing lots of people and replacing them with AI. And, you know, maybe that's the straw that broke the camel's back, but what we can say is they all had tons of tech debt. They were all doing the equivalent of like, paying their credit card bills by taking out another credit card and doing a balanced transfer. And like, eventually they had what our friends in the finance industry call a maturity mismatch on their plan of just like kicking their tech debt into the future. And then they all went tech bankrupt. - Right, and so it's funny, I immediately sort of thought of this as in shittification. And although it doesn't necessarily fit the mold as something like of like, you know, Google trying to just like squeeze its customers more and more and more. This is more just sort of a, yeah, as you said kind of an austerity measure turning to the new tools and then kind of, you know, it's almost like the mindset that Doge kind of made like so crystallized, right? Like we have the technology coming or we're gonna say we do, we're gonna apply this logic to the system. And, you know, hollow out all of the people doing the real work, the maintenance work because we don't need it for these systems anymore. You wrote a great blog about the Microsoft outage in particular. And I'll just read this one quote that jumped out at me. If you can't manage your servers online, you have no business incinerating tens of billions of stock buybacks, much less hundreds of billions to feed the AI money furnace. Did Microsoft need the tens of thousands of workers that threw away demonstrably so when they were on the payroll, Microsoft servers were up. Now they're gone and so are the servers. And you go on from there, but I think it's just like, the fact of the matter is we are seeing demonstrable impacts of sort of, you know, this way of thinking again, whether it's like AI that broke the system or not, it's as you're saying, it's the lack of attention to actually keeping them up and lack of care in the work forces that could do so, right? - Yeah, I mean, I think the thing that connects us within shittification is, in shittification is what you do. If you reckon that you won't face consequences for harming one or more groups of users of your service, it's, you know, this is Ed Zittron saying, you pigs are gonna whatever slop, I tell you you're gonna eat, right? That's the way that Ed describes us, right? And so, you know, I think that that one of the things that's characteristic of Doge is not just this real and also frightening belief that, oh well, invent the technology that will fix this tomorrow. The most haunting thing I read about Doge, I think was in a wired article and it was this guy from the NIH or a scientist from the NIH, begging them not to shut down long running cancer research studies and they were like, well, we don't need the studies 'cause AI will cure cancer, right? So there's that, right? But there's also this sense that like anyone who relies on the government shouldn't expect anything good. And this is the, you know, this is not the first person to point out that when you're a reactionary anti-government movement, that making the government worse when you control government serves your political goals, so, you know, these people were all anti-government extremists. And so, making the government suck is definitely within their purview. I think Microsoft is just a monopolist. I'm fact, it's not even controversial. They were convicted of having a monopoly. - Right, they are officially monopolists. - Yeah, I think they get to display the little plaque in the lobby. And so, you know, it's just GW Bushlet them off the hook 'cause they dragged the proceeding out for so long. So, yeah, this is like too big to care, right? This is just like we have circumstances that let us not care. And I do think also that in all of these cases, we do have this sort of prima facie case that they cut too deeply. And, you know, they can have all kinds of excuses. They can say, well, the workers we fired weren't doing the important work. They weren't related to the things that went down. It's like, okay, but you still don't have the managers to figure out which workers you should have hired to keep your servers from going down. You know how I know? Your servers went down. (laughing) And not just like went down. Like, outages happen for like an hour or two. But, I mean, this was days. This was days on end. I was actually, you know, I was working with a bunch of people who had to, you know, who were using Outlook and who were using Microsoft products. And they were like, we just cannot do anything right now. It's done like for not one, but two separate days. I don't know that I've seen something with that long of a tale for years. Yeah, it's pretty amazing. You know what it reminded me of, although it's not exactly the point they made, but JS10 and Clearser Red wine wrote this book against Teco LaGarkey. I think maybe you had them on. Did you have them on? - I had them on, yeah, but I did, I did blurb the book. - Yeah, that's right. You blurb the book. That's why I saw it. So you read the book too. So one of the points they make at the end. And for people who are thinking about reading the book, it's a good book all the way through. It's a bunch of war stories about unionizing tech shops. But the last 20% is basically pulling together all the lessons from all these union drives and going, what is it about tech labor that makes unionizing tech labor different from unionizing any other form of labor? And one of the points they make is that if you go on strike and you work in an auto assembly line, no cars come off the line. But if you make software and you go on strike, your software keeps running for a really long time. And so this really changes the dynamics of labor activism. There's a corollary to this though, which is that if your boss fires you, the consequences of firing you come a long way later. Because your code keeps running for a long time. And you know what it reminds me of? Is there's a whole class of policy problems that have consequence separated by effect by a long range, like privacy violations, right? Getting your privacy violated doesn't result in you getting swatted or having your wages lower and or whatever, right away. But like given enough time, enough data, enough breaches, everything becomes really, really terrible. So you can build up a lot of policy debt by like individually making bad choices by privacy, but also societally not building political will to do something about privacy until you've accumulated so much privacy debt that we're all living in though sort of in fiery consequences. Same with monopoly, like there are all these policy problems where causes separated by effect by a long way. And it reminds me of when I gave up smoking. So I was a smoker for most of my life. And actually now I've been a smoker for longer than I haven't been, but I was a smoker from 12 to 30. And even my wife, 'cause we were the only smokers in an event, right? And the reason we're still faithful to each other is we both gave up, we can't meet anyone anymore. And smoking, if I had gotten a tumor with my first cigarette, just a little one, that my last cigarette, right? I had cancer two years ago, I beat it, right? But I had cancer two years ago, 20 plus years after I quit smoking. I don't know if they're related, but I think it would be crazy to say there's no chance they're related, right? And when I had a cancer diagnosis, it really crystallized my feelings about my exposure to carcinogens, in a way that was hard to grasp when I was putting that first cigarette in my mouth. And in the same way, you fought, like Elon Musk walks into his data center after buying Twitter and starts pulling ethernet cables. Do you remember he did this? - Yeah. - At a server racks. And his SREs were good at their jobs. So you could pull down a whole server racks and Twitter wouldn't hiccup. But once you take all the redundant server racks offline, the first one that blows a power supply, takes Twitter down. And in the same way, so this is like this, it's a specific, he had fired a bunch of those employees. So, you know, I just, I think this is like a pernicious kind of problem broadly that, you know, we get this, this thing we're causing a factor separated by a lot of time and space. It serves scammers. It confounds policymakers. It trips us up. It lets bosses get into a lot of trouble. It, it, it makes for a really thorny political arguments. And so, you know, you put all those things together and, and you really have the makings of a very pathological situation. And really, you know, maybe like maturity and caution and probity and all the values we want from like a government or, or corporate management, like a big part of those are just like paying attention to those kinds of things and going like, you know, not to be a crazy person about here, move into a bunker. But what if this goes wrong? Like what's our cushion? What's our forward plan? You know, what caution do we have? How do we fail gracefully? And this is, I think it's a, it's a really good point to raise, especially sort of at this moment where we are at maybe the apex point in modern tech history of like not wanting to answer those questions. Just wanting to say, hey, look what we've got coming. AI is right around the corner. It's going to be able to solve everything. It's going to be able to do everything. It's kind of them. It's, I mean, you literally have guys like Sam Altman saying like we're, we're just around the corner from a genie Dario Amode and Anthropic is saying the like the age of miracles is coming thanks to AI all like promising like there's never been, I think, more of an incentive to just disregard all the stuff that's that's cut that's breaking right now and to say like just put all your money into this. This is the case that they're making. AI is going to be this, this by all, this be all end all sort of fix it all. And you know, they're out there making this big IPO push, but at the same time, you know, things are wobbly in AI world too, right? With the Jensen Huang ouster, with the sort of the new, all the questions swirling around the regulatory regime. But yeah, what, if you could speak a little bit to what role AI is playing in this, in this moment and sort of how you're thinking about the AI bubble given all of these sort of new ingredients into the stew. Yeah, I continue to try to kind of crystallize that my thinking about this. So I'm not like a rabbit ante. I have to say that like, I don't think technology has kudis. Like I think the answer to technology that was made by bad people for a bad purpose, but does something good is to like clone it in free and open source software and make those people broke and continue to use the technology. The best thing to do when you discover, you know, this is, this is a deep one for Linux heads. If you, if you go to the lab and the drawer is locked that has the paper tape for your printer driver, you should just get angry and spend the next 30 years cloning Unix and then put everyone who makes a commercial Unix out of business by doing that. Like that is the, that is the correct and reasonable response, not to go like AT&T or bad people in their war criminals and anything they've touched as bad and should never be used. So Unix is bad now. So I'm like, okay, with people going, oh, there's some useful things we can do with AI and, and the correct response to that is like fine. Let's like figure how to do it without torching the planet. Right. Well, the problem isn't just that isn't the technology. I would argue that problem is, is this ideology is this is sort of the bit the motor behind the gargantuan sum? So they're not saying, oh, there's some cool things you can do with this technology. They're literally saying, right. Yeah, exactly. We're going to teach somebody worse than we're guessing program that it's going to become God. So the problem is so buy in and that's it's God. So get your, you get your like investment share in God now before it's God and you can really, you know. So I'm both like, I'm, I'm mostly a materialist because I'm a good old lefty. And so I think that like people do these things for material reasons. And I think the material reason that people love AI, the investor class loves AI, the non ideological reason is that they think that a sales call where you show up at your prospect's office and say, hey, you know how like you have all these underlings who know how to do things that you don't know how to do and you have these ego shattering confrontations with them where you ask them to do things. And they tell you that you're really in ideas are like illegal, immoral, impossible will kill people. And you can't do them yourself. How would you like to fire 90% of them? But the rest of the, put the fear God in the remainder have an AI shit out all of your product ideas. And they can just bang them into shape without all of us nonsense. And that is especially true for a boss whose reports are people who are like statutorily or professionally required to tell them to fuck off. So like the boss of a lawyer or doctor, an accountant, right? Like, you know, they're just not allowed to break the law and they have like professional licensure and regulation that like requires them to tell you to fuck off when you tell them to break the law. And so I think that like you can be a materialist and just go, oh, that sales call is going to work really well. I don't know. It doesn't have to work, right? It just has to be plausible enough. And I have to have exited the market before anyone figures out that it doesn't work. But I also think that there's this ideological aspect because the thing that drives that makes this AI different from the thing we called AI the last time we had a thing we started calling something AI, which we were calling machine learning, but also AI is that the old machine learning that we used until about 15 years ago did the same stuff that this stuff does except that the first step was to make a model that described the world, not a model like an LLM, but like a model like a ship in a bottle. Like if you wanted an AI to figure out how to play chess better, you would start by just writing like a completely normal chess program that had like a software representation of a chessboard and chess pieces and the rules of chess. And then you'd use machine learning to iterate over it and come up with chess gambits. And the problem with that is, first of all, like all most programmers know how to play chess, but lots of programmers don't know how to, you know, how the pancreas works. And so if you want like to get a programmer to write software to help you figure out new therapeutic interventions for pancreatic cancer, you need to pair them with pancreatic cancer experts. And so that's like expensive and it's difficult. And then there's like a whole realm of things that we observe causal relationships with, but don't understand them. Like if you get sick and you go to the hospital tonight, in your hospital's pharmacopeia, are lots of medicines whose action is not understood entirely or even not at all. And we'll still prescribe those medicines, even though we don't know how they work, although we'd like to. And so the thing that LLM's brought to the picture, that this blockbuster paper, which is, it's a really interesting paper that people should read, whether you're a rabbit auntie or a rabbit pro or anyone in between called attention is all you need. It's the thing that started it, one the touring prize, one the Nobel. It's big insight was, what if instead of describing the world to the computer, we just took every observable related fact about the world and gave it to all the computers that they'll let us use. And then ask the computer to just figure it out without knowing what those tokens, right? Like this, these two blobs in a vector array, what they represent. So you take every conceivable chess move that's ever been played and its relationship to all the other chess moves and all those games. And then you just say like, if this chess move happens, what's the next chess move? Like what's the likeliest next chess move? Or it's not really likely is because there's some some genuinely interesting math that goes in there. But like what what chess move is a good chess move to put next. Without knowing what a chess move is. And boy, boy, did this work really well? Like like work better than anyone who paid attention to machine learning, thought it would, right? Like this is a real like genuine breakthrough. And if it's a a breakthrough of any kind, it's almost a philosophical one because it works as well as it does because it turns out there's more statistical smoothness in the universe than we would have thought. Like things things in the future are a lot more like things of the past conversations are a lot more like conversations that have happened sentences are a lot more like sentences that you can really predict what words come next in a sentence based on all the sentences without knowing what a sentence or word is. The problem of course is that if you ask a computer to play chess this way, it sometimes puts a piece on a square word already has a piece because it doesn't know what a chess board is, right? It's just making statistical predictions. Now we call that a hallucination, but it's just like a function of this kind of guessing. So the the ideology that steps in here, so you have this material thing which is we can sell a lot of employee eliminating software to bosses who hate their employees. And then the ideological thing is I a capital allocator billionaire and continuously thwarted by experts and there has never been an exchange market for dollars and expertise, right? I mean there's an employment market, but if you like you have a crank idea about how to live forever by doing something to your pancreas because you're that weirdo who injects his kids blood, right? And no pancreatic expert wants to work with you on this. That's it, right? Like you can't there's an amount of money that makes new pancreatic experts materialize who want to work with you. But this is a mechanism for trading resources for expertise and a reliable way assuming it works, right? Like you can just take brute force that knows nothing ignorance and you can refine it into knowledge. Without an expert, and like that's a thing we've just never had before and I can see why it would make someone whose biggest frustration is being thwarted, really excited about pledging their fortune and their immortal soul to it. And I think that when you have a scam or swindle, that's being run by a mix of true believers and scammers, right, when the White House has Trump and Pence in it, right, a scammer and a true believer, they go farther than either one would go on their own. And so can the scam pay off, right? Like, you know, the funny thing about the Ponzi's hands, did you mean Vance or Pence? Of, uh, Pence, Pence, Oh, Pence, the original, yeah, yeah, no, no, Pence doesn't believe in anything. The reason the second White House has failed is just scammers and scammers, right? The first Trump White House was able to do things that it was able to do because it was scammers and true believers. And whatever was a bridge too far for Trump was ideologically consistent for Pence. Like there's lots of stuff Trump just doesn't give a shit about because he doesn't give a shit about anything about himself, whereas Pence really believes in a bunch of really unhinged things and will do anything to make them happen, whereas there are a lot of things Pence won't do because he believes in things and Trump believes in nothing. So there's nothing he won't do if it advances his own agenda. And so you put them both together and it's like so much more powerful. And I think that's the AI bubble. And so now we have this question, right? Can we make, can the scam pay off? Because like one of the weird things about, about, uh, Ponzi, the original Ponzi is the original Ponzi scheme might have worked. He figured out how to buy these things called international reply postage coupons that were worth like 40 cents for 30 cents. And if you can buy things that are worth 40 cents for 30 cents and you can find a market to sell them, then you can effectively make infinity money. But it was just so much easier to get people to put more money into the scheme than it was to go buy these international reply postage coupons and sell them. That he just gave up on the scheme and focused on the scam. So can we make the scheme of AI pay off or is it just a scam? And so for the scheme to pay off, what you have to do is replace a lot of skilled labor with chatbots. And the problem is that because of the way that this theory-free inference works, where you don't tell it what the world is and just give it tokens and ask it to make guesses, is that it makes errors that are statistically close to being correct as it's possible to be while still being wrong. And so for AI to let you fire a bunch of workers without sacrificing quality, for AI to let you fire a bunch of workers, have the remaining workers be the humans in the loop who check the errors. Those errors would have to be spotable by the human. If you fire 9/10 of the radiologists and have the remaining radiologists do 10 times as much, you know, radiological diagnostics by signing their name at the bottom of the radiology report generated by the chatbots, you better hope that when the chatbot makes a mistake, it's really obvious. They're like, "Well, you know, I looked at this chest X-ray and you've got the Starship Enterprise in your chest," right? And not like this thing that most of the time looks like a tumor, but this time didn't isn't a tumor, which is going to be the kind of thing that's really hard to spot at scale. Now could you give a radiologist a chatbot that examines all of their work, and twice a day says, "Take another look at that one, and one day a week is right, and you save one person's life a week," I totally believe we could do that with AI. You cannot pay off the cost of developing theory-free inference by making radiology more expensive and slightly better. Right. Exactly. That's the impossibility, right? I know it's a long walk-up to that point, but that's why AI is a scam and not a scheme and can't be made back into a scheme. Now maybe when they go bankrupt and we buy all their assets really cheaply, and we just do normal shit with it, maybe it can be a scheme, maybe it can just be that thing that asks the radiologist to check their work every now and again. It's just an "are you sure" dialogue box for your, you know, when you close a document or whatever, right? Right. Right. But as you're saying, the incentives are all misaligned. The promise is mass labor savings. The promise is less, you know, intrusive and less powerful labor. The promise is automation, not, you know, not augmentation, not this temper. So that's what's motivating, as you're saying, the bubble. And I think, you know, we haven't looked at the cause and like the nitty-gritty of each of these failures that we started, you know, talking about. But with Microsoft, it seems pretty clear as you point out that, you know, like this is something that could have solved, right? Yeah. With FAA. Yeah. I'll go ahead. Yeah. I was going to say, you know, in automation theory, well, I, maybe I've even read this from you, but in automation theory, labor-driven automation historically is in service to improve quality. Like when a skilled crafts person picks up a tool, it's because they think they'd let them do a better job. And capital-driven automation is to improve throughput. And when the capital has market power, it's to improve throughput at the expense equality, right? You pick easily whatever slot we give you. So like the really interesting kind of historic comparison, because AI bros love to compare AI to the web. They say, oh, the web lost money, but now it's profitable. So there's lots of ways in which that's not true. But one of the most interesting distinctions between the web and AI can be found if you go read the business press of the odds in the late 90s, which is full of articles about how CEOs can cope with all of these young workers who demand the use of the web on the job site. And when you read those journals today, Harvard Business Review or Business 2.0 or whatever, they are full of articles about how bosses will cope with workers who refuse to use AI. So if you had any doubt about whether this was capital or labor-driven uptake, like that answers the question. Yeah. Yeah. That's such a great point. I mean, that's such a great point, because again, it just falls to sort of the worker. If the worker is actually properly empowered to make the decisions about how this can augment their skill set, or then they will do so if it's going to complicate, add more labor hours, more work, then they're going to have a problem with that. And that's what we're seeing right now with AI, right, in survey after survey, workplace survey after workplace survey, you might have seen the insurance industry, like the insurance claims adjusters. Yeah, sure. Nobody hates AI more than them, because they have this expertise, but they also have this, they don't rise, I think, to the level that you're talking about where actually they might have the power to tell them, tell their bosses to fuck off, because they don't have that level of power. Instead, they're trapped right in that the worst layer where the AI is creating plausibly correct, but wrong, often information that they then are fully accountable for. Yeah, I think it's even worse than that, because of course, because they don't have a lot of labor power, one of the things their bosses have pushed them to do is as many denials as possible. Yeah. And so there is this very adversarial relationship between insurance claims adjusters and the insured. And there are scammers out there too, of course, obviously there's scammers. My friend, Catherine Moronex says every complex ecosystem has parasites, right? There's always going to be scammers, but if you just like legitimately, like, you know, the last time I had an airline lose my bags and I had to call my travel insurer, my attitude going into this is they're going to try to not pay me the money they owe me, right? Like everything that this is just, this is about them finding a reason to say no, even though I paid them. And so what do people do? They get AI to write them insurance claims. Yeah. Yeah. And so these insurance adjusters are being crushed by appeals, by claims, by lengthy claims, by lengthy over-documented claims, by claims that make reference to hallucinated insurance regulations and hallucinated clauses and insurance policies. And so yeah, like, I get why they're being told that they need to use AI to fix it. But like, there's this joke that comes from Newfoundland, which is, I see a Paris Marxist book on the shelf behind you. It's where Paris is from. There's this joke that comes from Newfoundland whose punchline is if you wanted to get there, I wouldn't start from here, right? Yeah. You know, I understand why they're like, well, we don't have any air traffic controllers in the pipeline and we have a staffing problem. Yeah. Let's see. Let's use AI. Well, okay, sure, but the reason you don't have any air traffic controllers in the pipeline and you have a labor shortage is because for 25 years, this is the, you know, 50 years air traffic control has been governed by people whose answer to, oh, there aren't enough air traffic regulators is, let's find a computer that can maybe do the job and hope no planes crash. Right. And most specifically, just last year, fired hundreds of them just because they could, right? There was 400, I think, FAA employees fired by Doge and one of the early, you know, moves of the Trump administration. So we, let's, yeah, let's, let's tie all this together. So we have this moment right where sort of the logic of AI is reducing capacity or interest in generating more sort of generative capacity of employment in all these different fields. It's frustrating workers just about everywhere. you know, that is, you know, there's some fields that have gone, okay, we figured out how to use this, but I think more often than not, because of this sort of, you know, top down in position of AI by management. It's creating, you know, more tension than productivity. And yet, it's IPO season, right? So they're out there making the pitch for, you know, they're trillion dollar companies. You know, where, where, where, where, where, where, where does this go? How long can this sort of, it's going to cure cancer up here? Just rumbling, dissonant kind of trouble brewing everybody. And I think it's important to note that whether or not, again, it was AI's fault when the planes were grounded, everybody was just immediately assumed it was AI, because that's their experience broadly, right? For one, we've been told it's going rogue all the time and it's bringing down systems number two. It's just making everybody's lives a little bit worse. They see slop on the internet. They see, you know, they hear, you know, they're boss talking about efficiency and maybe they see their colleagues getting laid off or they're not hiring as they should. It's just creating a layer of unpleasantness. So that's the experience of AI on the ground. Again, cure cancer up here, people are getting rich. What, what, what is going on? How long can this, how long can this, this formation hold? So the how long question's really interesting. I have two poles that I kind of oscillate between. So one is this bedrock of finance lore. It's called Stein's Law. And it says anything that can't go on forever eventually stops, right? And, you know, hard to argue with, right? And I think we see it around us all the time. We see these, like you have these phase changes where if you think about like an avalanche, right? So an avalanche is a set of very small, almost undetectable destabilizations where things tremor and shift and become a little more unstable. And eventually you get to the point where one more instability added to the system doesn't cause a minute almost undetectable shift. And so anything that can't go on forever eventually stops, you can't just keep dropping snowflakes onto the unstable mountain side, eventually the combination of heat, rain, temperature fluctuations, more snow is going to make it all fall, right? And so I think this is true of the finances, the AI sector, which are brittle from several different directions, right? We learned a new source of Brittleness, which is that the people running these firms are insatiably horny and vulnerable to Chinese spies. Like, you know, there's a baroque data center financing deals that are being financed by Emirates who suddenly need to rebuild their liquid natural gas terminals because Trump decided to get a rant to bomb them into smithereens, right? So that could be the thing that does it. And then there's these IPOs, right? So I'm sure I heard you cover this, but you know, the Musk IPO was such an amazing swindle in that he got these sweetheart deals from NASDAQ. He was originally gonna get them from S&P. And S&P, eventually like when fund managers came along and said this is like, this does not comport with the probity we need from a major stock index. They, and they said that forcefully enough, S&P walked away from it and walked away from the Musk IPO. But NASDAQ gave him the steal where he could enter the index without being seasoned by having a couple of quarters of profitability and growth, which is normally when your stock goes IPO, even if it's in the top 100 of the NASDAQ, it's not in the NASDAQ 100 until like it's, there's been some time for the system to shake out for bugs to show up for people who know that it's a swindle to like short your stock and crash the price or whatever. Like it has to show that it's got some staying power 'cause of course the financial advice that you got and I got and everyone who's ever done that consumer finance class in high school that's mandatory has gotten is take a certain percentage of your earnings and stick it in an index fund and forget about it until you're 65, right? Like that's the advice Warren Buffett gives people, right? So everyone's savings. Everyone who doesn't wanna starve to death or be homeless in their old age is in one of these index funds. So it really matters if the index funds are stable. So NASDAQ is like, now you can go into the index fund on day one and then they also said your insiders can sell their shares on day one. 'Cause this is the other kind of, you talked about misaligned incentives. If you know that your stock is a pile of shit and you can sell it on day one, you go public and you run away as fast as you can. Whereas if you have to hang in for 90 days before you can sell any stock, then you at least have to make the scam good enough to last for three months, right? And by allowing musks insiders, right? The biggest shareholders in SpaceX were people who gave money to Bill Groc and to buy Twitter and they converted their debt to equity in SpaceX. SpaceX was like this, matrioshka that is outer layer, outer layer was like, you know, totally normal beltway band that overcharges Uncle Sam for services that the government should provide to itself. And then inside of that were two piles of shit nested inside of each other. Groc, which is the world's most efficient money furnace and Twitter, the stupidest tech acquisition go private in, you know, capital market history. And so what Musk did was he sold almost no SpaceX shares that were issued by SpaceX. And so you had fund managers who had to buy SpaceX shares at any price legally. If SpaceX is 3% of the NASDAQ 100 and you are a NASDAQ 100 fund manager, 3% of your portfolio has to be SpaceX shares and you have to buy them at any price they're being sold at, Musk does not make any shares available but his insiders can sell. So we end up being the exit liquidity for Emirates and Silicon Valley oligarchs. Now, this is so bad. You want to talk about something that is going to make people really angry. I was just writing today about the Georgia Senate race where people couldn't afford hamburgers and the, or not Georgia, Mississippi. - Mississippi. - And the Mississippi Senator said there are other proteins, you know. And just like suddenly Mississippi, like, you know, 25% of the white vote went to the black Quays Eye Socialist candidate in Mississippi because they were like, fuck off with your own not allowed to eat hamburgers anymore. - Right, pick another senator, right? - Yeah, imagine what happens when you've lost your life savings. In fact, we don't have to imagine it. You know, I go through my old blog headlines every morning from this day, but like 25, 20, 15, 10, 5, and one year ago. So 10 years ago, we were dealing with the aftermath of the Trump election, the first Trump election. And there were some statisticians who found that the biggest predictor of a precinct switching its vote to Trump was how many foreclosures that it had, right? So, you know, 'cause that was Obama's policy, phone and ways for the banks, let the foreclosures happen. - Right, you get voting, you're angry, you've lost your house, lost your savings, yeah. - So imagine the political blowback, right? From anthropic, going public, crashing, people losing their retirement savings because they put their money in an index fund. They never went out and said, "I want to buy an anthropic." - I mean, that's the thing that has to be underlined again. Like people just like, there are people who may not even know that they have SpaceX stock right now, because it just is in the index fund. And then now, and anthropic, and then presumably next year, open AI too, that's the, yeah. - So a failed anthropic IPO, right? Like if NASA says, "Sorry, no sweetheart deal for you," because now the Democratic Congress is pulling the SEC chairman in and saying, like, you need to exercise some probity here. We can't afford to, you know, deprive American savers of their life savings by giving them exposure to stock volatility when they're buying index funds. No more of these sweetheart deals. Musk got his deal, nobody else gets this deal. That could be, that could like really fuck up the anthropic IPO, right? And then that could be the thing that spooks the market and starts to see margin calls and starts to see the lenders who loan money to all the big tech giants, you know, not roll over their loans the next time they come due. And, you know, or you could see a cascade where further down the people who loan the money for the data centers walk away. And then with the data center bubble imploding, then that prompts the lenders to the tech giants to walk away. Like, it's, the thing about a bubble is all the parts of the bubble are fragile. And you only have to get the pin through any point in the bubble for the whole bubble to burst. And the number of pins pressing in on the bubble is so large. That's what, that's what it feels like to me right now. And I, you know, you know, you and I are in this business. We know better than to say the bubble's about to burst 'cause every time in the, you know, you've said, oh, one of, you know, I have, I have also said like, there's no way that this can go on much longer. And then another year goes, oh, well, that's the other poll, right? So I said, one poll is Stein's Law. Anything they can't go on forever eventually stops. The other poll, John Kenneth Galbreath, The market. can remain irrational longer than you can remain solvent. Right. We just why we can't time this. We can't and we and we every time we've tried to we've tried to do so before it's, you know, it has carried on, but and yet and yet and yet. Like right now it does seem as you say with with in videos sort of in this deeply uncertain state. And then that sort of rippling out to the Neo cloud companies that are losing stock value, maybe even more than Nvidia in some cases. And with that with the circular financing that Nvidia was kind of propping up or big question is whether they can continue to do that for the broader AI startup ecosystem incoming regulation. Who knows who knows? But it, yeah, remember, Jensen Wang says it's not circular financing. It's cyclical financing. Right. It's, yeah, it's a cyclical financing. My bad, Jensen. Well, we now know that he does have a certain affinity for, I guess, cyclical, cylindrical shapes, perhaps, right. Did you, did you see Sean Duffy when they, when the reporter said, so the air traffic control crash was not caused by AI. This is the transport secretary used to be Fox News host. Yeah, he said, I think you mean SI. Yeah, I think that was a Mark Wayne Mullin who said Mark Wayne Mullin. That's it. Yeah, it was Mark Wayne Mullin. Yeah, there are thoughts that, you know, this could be an AI related error or issue. We call it SI around here. Remember SI. Jesus, first of all, I know, I know. But again, like to me, that really sort of underlines this whole thing. People are saying, is this huge problem, you know, that's, that's, that's affecting Americans, millions of Americans is, is the problem AI? And not only dodges the question and doesn't interrogate what's actually going on. He says, no, no, we call that SI super intelligence, right? Right. I'd say, you know, I'm trying to think of a, I'm trying to think of a joke that fits with this, but you know, it ends with like it starts with, you know, who killed my parents and all the other people I live with. And it ends with, I think you mean whom exactly or, yeah, or just like, no, think about like, can we get, can let's remember, we're thinking about the super power. Again, you're, it's not, it's, it's a different, you know, it's not a technical, you know, it's not, it's not a technical system. It's this, it's, it's this, almost this ideology, this ideological layer, right? Finally, I, so, first of all, we're all, you know, so glad that, that, that, that you beat cancer that it's gone. And I, as somebody who also, I, I, I didn't smoke until I was 30, maybe closer to like 28, maybe it was 30, I don't know. I smoked for a long time too and, and quit. Well, you know, you, you beat it, right? You beat, fortunately, things worked out. You got good medical care. You got, well, I used AI. It told me to use, but not, not everybody's gonna, gonna, gonna beat it, right? And our medical system is currently buckling under the weight of all the people that need care and some people can't afford it because of the, the, the security incentives of the system. And so right now, are we, I just worry that we're baking in so many of these failure points, things that we're not recognizing and that when this bubble burst, I think, you know, there, there could be a lot of damage that we're not catching now, that we're gonna have to own up to later. I mean, how worried about that, are you? Yeah, well, so there's sort of, the two ends of that. So one is process knowledge, right? So process knowledge is all the stuff that's not IP that keeps the business running. And we talk about IP all the time because it's like the part of how the business operates that you can write down, put in a filing cabinet and sell with the business. So of course, markets really value IP, but process knowledge, so it's sometimes called tacit knowledge, but tacit knowledge usually refers to things that one worker knows and process knowledge is like throughout the process. So it's like workers within a division across divisions and affirm and then across firms in a sector. And it's just all this kind of lore, almost like, okay, we're, you know, we have this machine at jam sometimes, there's like the one person who knows how to fix it, but also if that person's on vacation, the person who used to fix it who retired last year, I have their cell phone number and they'll fix it for 50 bucks. And that's the difference between like a factory that runs and a factory that shuts down, right? And process knowledge can't be rebuilt, like a recovered rather, it has to be like regrown and it takes years and process knowledge is so important to affirm that you can actually shut down a firm with a labor action called a work to rule. And the work to rule is when you show up and just pretend you have no process knowledge, you only follow procedures and you never use any of this tacit knowledge that you have. And then firms just like literally cease to function when they operate this way. And so I think we're going to fire a lot of people. First of all, we're going to fire them replacing with chatbots that don't work. And then we're going to have this environment, this economic cataclysm and we're going to fire a lot more people and we're going to lose process knowledge and it won't come back like one of the reasons the chips act didn't yield to dividend within the four years of Biden is that like figuring out how to make chips requires a lot of lore that has to be built up. So that's the that's like basically we're going to lose the old people in their knowledge because the way you acquire process knowledge is by working next to someone who has it. And you kind of was most set by observing them. And by asking them questions, no one thought to ask and they give you answers, they didn't know they had. Let me give you one example of process knowledge that just blows my mind. So there was a chip foundry and they had a huge batch of chips and they failed QA and they could only run at half their nominal clock speed. And the person in charge of this knew that they had a customer where there was a division that had been trying to make a low end computer, but that they couldn't get it through the business. No one else like they thought their idea was any good. So they called them up and said, do you think you get your low end computer built if I could sell you the chips for peanuts? They said, yeah, I think we could. Can you sell me those chips? So they sold these low end chips. The computer launch was called the MacBook Neo. It was the most successful launch apples had in 10 years. It was so successful. They used up all the defective chips and now they're making new ones, right? So that's what we vaporize in a crash. And it's what we vaporize through mass layoffs. And then at the bottom end, you have all the people who are going to be disillusioned, right? Who are never going to enter the economy because they're entering an economy that is in collapse, right? You know, think of like people who never finished med school in Greece during the crisis because this was the era in which they were, you know, like then they were saying, oh, you can't have cancer drugs anymore because your government owes too much money to German bondholders. They were just dying of cancer, like who trains to be an oncologist under those conditions. So you have this like lost generation of oncologists. Your pipeline is empty, right? You can't just like, again, you can't just like make a new oncologist from scratch, right? You need, like they need, you can't conjure them up when you need them. It's like a 12, 15 year process to make an oncologist. So I'm worried that like we're going to wipe out both ends of this and that we're going to go into this poly crisis of environmental crisis and refugee crises, geopolitical crises, empires that are, you know, self-imulating. And we're going to do it in an era of sharply reduced productivity and sharply reduced administrative capacity. And so we're going to be like fighting the fight of our lives with one hand type behind our back. Yeah, sharply reduced number of skilled, you know, workers and yeah, yeah, I, you know, I think that that's, it is, it is a real, I mean, you just look again, I think a lot of the question is I, you know, you say sometimes it's better for a bubble, you know, to bridge, it's always better for the bubble to burst sooner. You know, you were talking about that on your, you know, book tour for, yeah, for your latest book. And that one reason I hope that is you look at, you know, the way that, you know, these entry level, you know, employees of people leaving college looking to sort of the job market, looking to sort of get their careers and their lives started and they're kind of, you know, being told like, oh, AI can do a lot of these jobs or AI can do. And so they're, they're not, they're not just losing an opportunity to sort of build those skills, but they're losing like an interest in a connection to the broader project and the prospect of, you know, of, of having those lives that could be rich in fulfilling and doing these skills and starving the pipelines. So like again, starving the pipe, you know, if we, if like eventually radiologists are going to retire. And if there's no radiologists in the pipeline, because we told them that AI can do the radiologist job, we will end up in a position where the only thing we can do is have our dwindling number of radiologists look at a hundred times more chest X rays than they ever have before and try like hell to spot these errors that are statistically close to being correct as possible, while still being wrong with the AI. Yeah, and we just won't be like we won't be able to, you know, insert enough humans to make the difference. It's self-fulfilling prophecy. If you tell mathematicians that AI can do math, then we won't have mathematicians, because who's going to enter the field? Yeah, exactly. Well, again, every time that I say, like, well, I can't get, can't bottom out much further, you know, if we reach a new, a new low and a new layer of sort of fractures spread across the ecosystem, but, you know, I do think, I do want to end if you have another second, yeah, with just a, you know, some hopeful ideas, because I know that. Yeah, so here's my hope. right so my arch nemesis and yours whether you know it is this guy called Milton Friedman who's Satan took to hell in the early 2000s and Milton Friedman was the like the designer and architect of this neoliberal system we live in and for years he was correctly viewed as a crank right during the great society years what the French called the 30 glorious years right the post-war years when we had broadly share prosperity big social safety net tax high and high taxes at the top of the income distribution and so on people thought he was nuts they were like Milton people like you know their grandparents not starving to death they like their kids going to university they like having labor rights they like not breathing poison there how are you gonna convince them otherwise and what Milton Friedman used to say is that in times of crises ideas move from the periphery to the center and our job is to keep ideas lying around so that when the crisis strikes we our ideas can move to the center and during the oil shock of the 1970s Friedman was able to move his ideas to the center and wasn't Reagan it was Carter who took them up I mean Reagan did it in a way that made Carter look like an amateur but Carter started it as Democrats who are like well I guess right I guess this is the idea that's like at the periphery we tried everything else everything's breaking down what are we going to do so our idea our job is to keep good ideas lying around to understand that you know there is labor power can solve this that well-free distribution can solve this that wealth predestribution the prevention of the accumulation of wealth can solve this that's our job right is to have the good ideas lying around that technology should serve people that we should seize the means of computation right and that and and when the crisis strikes we should be there and we should say like I mean what is the midterm election if not good ideas moving from the periphery to the center you know if Trump is a factory for producing crises right right yeah so we just need to have an anti-shock doctrine kind of yeah yeah yeah well we can we can keep we can workshop it it's I agree I could not agree more I do think that I actually one thing that I that I also sort of keep in mind at these moments and I and I like to say is there is I think a lot of there is a lot of reason to be hopeful in this moment the like the sort of the fragility of a lot of these systems there's a I do think you're absolutely right spaces will be opening up in ways that we're not anticipating or understanding quite yet and there's going to be real opportunity for you know for reform maybe radical reform in the way that we you know conduct and construct and maintain our technological systems in the way that we sort of you know construct policy so yeah I think of the bubble is going to be bad when it bursts but you know there's going to be pain there's going to be a lot of pain no matter what but the only thing worse than that happening is us not capitalizing on it to make something better is for all of that suffering to go to waste right same with rising seas right we put too many terms into the ocean to avert rising seas like the second law of thermodynamics isn't up for grabs we won't make the seas colder again we're losing a significant amount of the polar caps that's happening what we do about it is up to us do we welcome in the refugees or do we you know do eco-fascism it's our choice right we we can see that crisis coming and you know like utopia is not a world in which things never go wrong utopia is a world in which we address our failures gracefully and make a better world well thanks again Corey so Corey you're probably you know everybody knows where to find Corey online it's the pluralistic is the blog he's on social media he's got books everywhere so let me just what where what should people be looking for yeah in terms of the greater Corey doctoral universe right now so so an April there's a graphic novel out from for a second Macmillan it's a middle grades graphic novel adapted from my short story on authorized bread a heroic story about kids y'all breaking appliances fun story it's in that's in the radicalized group of fellas yeah and then in August there's a book Geopolitics book for adults all for also from Macmillan from for our stress and shiru called the post-American internet about the internet we can build now that Trump has blown up all the things that kept us from wanting to make Americans angry now we can make them angry because what's going to what are they going to do put tariffs on us and and then next winter either at Christmas or maybe early in the new year of 2028 the graphic novel and shittification is coming out right and the current book the most recently published is the reverse centaurs guide to life after AI reverse guide to life after AI on New York Times bestseller hey there you go all right well keep I'm coming sounds like you you plan on it all right you sure it's good nice talking all right that's it for today thank you everybody for dropping by tuning in thank you to Corey doctoro it's always a pleasure thank you Ryan Otis our esteemed producer happy to be here and you're welcome well done I love putting you on the spot it's one of the things I look forward to every week so so thank you for that thank you as well to Kate Osborne Ryan Hall and current Shadme and once again to the listeners if you like this stuff please go out there review subscribe up vote I don't know all that stuff and until next time keep putting sanders on cheers

Podcast Summary

Key Points:

  1. Digital systems like Microsoft Azure, FAA, and banks are suffering massive outages due to aging infrastructure, poor maintenance, and a lack of skilled human oversight.
  2. These failures stem from a pattern of corporate austerity—firing experienced workers and replacing them with AI—leading to brittle, unresilient systems prone to collapse.
  3. The ideology behind AI automation prioritizes efficiency over resilience, often cutting staff who understand complex systems, thereby removing critical human knowledge and fail-safes.
  4. AI is not the direct cause of outages, but its promotion as a panacea masks deeper systemic flaws, including misaligned incentives, tech debt, and the erosion of tacit knowledge.
  5. The current AI boom is fueled by ideological enthusiasm and short-term financial gains, not long-term stability or societal benefit, creating a dangerous bubble.
  6. The "inshitification" phenomenon reflects a broader collapse of trust in digital infrastructure, where technical failures are blamed on AI while real causes—like poor management and labor cuts—are ignored.
  7. Historical patterns, such as Stein’s Law and Galbraith’s observation, suggest that unsustainable systems will eventually fail, and the current tech landscape is highly vulnerable to cascading collapse.
  8. If the AI bubble bursts, it could trigger a systemic crisis, destroying process knowledge, displacing workers, and collapsing key industries like healthcare and manufacturing.

Summary:

The recent wave of global tech outages—spanning Microsoft, the FAA, major banks, and airlines—exposes a deep systemic crisis in digital infrastructure. These failures are not due to AI malfunction but are rooted in decades of corporate austerity, where skilled workers are fired and replaced with AI, leading to brittle, outdated systems. Early digital systems were built on complex, custom code with layers of technical debt, and when merged through mergers and acquisitions, they become vulnerable to collapse.

The erosion of human expertise—what researchers call "process knowledge"—means systems lack resilience and fail when stressed. A major factor is the ideology of efficiency over sustainability, where companies prioritize cost-cutting over maintaining stable operations. This has been exacerbated by the rapid push for AI adoption, especially in sectors like finance and healthcare, where AI is promoted as a miracle solution despite evidence of poor performance and high error rates.

Critics argue that AI is not solving problems but replacing human judgment with flawed, statistically plausible hallucinations. The current situation reflects a dangerous bubble built on ideological fervor, misaligned incentives, and unchecked automation. If this bubble bursts—triggered by failed IPOs, regulatory resistance, or financial instability—it could lead to widespread job loss, the loss of critical expertise, and a long-term decline in essential services.

The crisis is not just technological but social, undermining trust in both institutions and the promises of AI. As seen in the dismissal of AI-related failures and the rebranding of "AI" as "super intelligence" (SI), the narrative is being manipulated to avoid accountability. Ultimately, the collapse of these systems is a symptom of deeper failures in management, labor practices, and the belief that technology alone can solve complex human problems.

The coming years may reveal whether this technological overreach leads to a sustainable future or a period of widespread instability and loss.

FAQs

The outages were likely due to systemic vulnerabilities in aging, complex digital infrastructure, not AI. These systems were built with outdated software, layered with mergers and redundant processes, making them brittle and prone to failure when key components are removed or disrupted.

No, AI is not directly responsible. The outages were the result of technical failures in systems that lack redundancy and skilled maintenance. While AI is being heavily promoted, the actual failures stem from poor system design and lack of human oversight.

Inshitification describes a process where systems become unstable due to austerity-driven decisions—like firing experienced workers and replacing them with AI—leading to a loss of critical knowledge and resilience in complex digital systems.

Firms are replacing workers with AI to cut costs and boost efficiency, but this ignores the value of human expertise and tacit knowledge. The result is brittle systems that lack resilience and are prone to catastrophic failure.

Mergers often combine incompatible, old, and custom-built systems with layers of complexity. When redundant systems are merged and staff are cut, the resulting systems lose critical maintenance knowledge and become highly vulnerable to failure.

No, AI currently cannot reliably replace human workers in complex, high-stakes fields. AI often produces plausible but incorrect results (hallucinations), and critical decisions require human judgment, especially when errors go uncaught.

Chat with AI

Loading...

Pro features

Go deeper with this episode

Unlock creator-grade tools that turn any transcript into show notes and subtitle files.